Accurate depth sensing by filtering IR cost on the guide of RGB information

Abstract

In this paper, we propose a depth-sensing method that combines the raw IR speckle pattern and RGB information. IR and RGB data are acquired from the same viewpoint. Matching costs between the realtime speckle pattern and the reference are computed with the IR data. Different from conventional methods, the RGB information is utilized to filter the matching costs. The depth at each pixel is then estimated by choosing the best match with the filtered costs. The proposed system can make use of the RGB information to achieve more accurate depth maps than the state-of-the-art, speckle-based depth-sensing methods.

Static hand gesture recognition based on finger root-center-angle and length weighted Mahalanobis distance

“…In contrast, the emerging depth images generated by depth cameras [10][11][12][13] are much more insensitive to the background. Hence many hand gesture recognition systems using depth camera have been proposed.…”